Assessing surprise for autonomous vehicles

Patent metadata

Field	Value
Publication number	US-11447142-B1
Application number	US-201916414176-A
Country	US
Kind code	B1
Filing date	May 16, 2019
Priority date	May 16, 2019
Publication date	Sep 20, 2022
Grant date	Sep 20, 2022

Abstract

Aspects of the disclosure provide for controlling an autonomous vehicle. For instance, a first probability distribution may be generated for the vehicle at a first future point in time using a generative model for predicting expected behaviors of objects and a set of characteristics for the vehicle at an initial time expected to be perceived by an observer. Planning system software of the vehicle may be used to generate a trajectory for the vehicle to follow. A second probability distribution may be generated for a second future point in time using the generative model based on the trajectory and a set of characteristics for the vehicle at the first future point expected to be perceived by the observer. A surprise assessment may be generated by comparing the first probability distribution to the second probability distribution. The vehicle may be controlled based on the surprise assessment.

First claim

The invention claimed is:

1. A method of controlling an autonomous vehicle, the method comprising:
generating, for an initial time using a generative model that predicts vehicle states from the perspective of an observer of the vehicle, a first plurality of predicted vehicle states of the vehicle at a future point in time;
obtaining, from a vehicle planning system of the vehicle, a first candidate trajectory for the vehicle to follow, the first candidate trajectory being one of a plurality of candidate trajectories to be evaluated;
generating, by the generative model, a second plurality of predicted vehicle states based on the first candidate trajectory, the second plurality of predicted vehicle states representing predicted vehicle states at the future point in time assuming that the vehicle has initiated the first candidate trajectory generated by the vehicle planning subsystem, wherein the second plurality of predicted vehicle states are based on a set of characteristics for the vehicle after initiating the first candidate trajectory;
generating a measure of observer surprise for the observer of the vehicle including comparing the first plurality of predicted vehicle states to the second plurality of predicted vehicle states based on the first candidate trajectory, the measure of observer surprise representing how surprising the vehicle following the first candidate trajectory would be to the observer; and
based on the measure of observer surprise, causing the vehicle to select and follow a different candidate trajectory from the plurality of candidate trajectories.

2. The method of claim 1 , wherein the generative model is trained to generate predicted vehicle states from the perspective of a passenger of the vehicle.

3. The method of claim 1 , wherein the generative model is trained to generate predicted vehicle states from the perspective of another vehicle.

4. The method of claim 1 , wherein the generative model is trained to generate predicted vehicle states from the perspective of a pedestrian.

5. The method of claim 1 , wherein the generative model is trained to generate predicted vehicle states from the perspective of a bicyclist.

6. The method of claim 1 , wherein comparing the first plurality of predicted vehicle states to the second plurality of predicted vehicle states comprises determining a Kullback-Leibler divergence between a first probability distribution representing the first plurality of predicted vehicle states and a second probability distribution representing the second plurality of predicted vehicle states.

7. The method of claim 1 , wherein the first candidate trajectory includes a geometry component and a speed profile that define a future behavior of the vehicle, and wherein the measure of observer surprise corresponds to how surprising the future behavior is to the observer.

8. The method of claim 1 , wherein the observer is a road user, and the method further comprises:
receiving sensor data identifying characteristics of the observer; and
using the generative model to predict a future behavior of the road user, and wherein generating the second plurality of predicted vehicle states is further based on the predicted future behavior of the road user.

9. The method of claim 1 , further comprising:
using the vehicle planning system to generate the different candidate trajectory for the vehicle; and
determining a second measure of observer surprise for the different candidate trajectory, and wherein causing the vehicle to select the different candidate trajectory comprises selecting a candidate trajectory having a lower measure of observer surprise.

10. The method of claim 1 , wherein the vehicle planning system is configured to continually generate candidate trajectories until generating a candidate trajectory that does not exceed a threshold measure of observer surprise.

11. A system comprising:
one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:
generating, for an initial time using a generative model that predicts vehicle states from the perspective of an observer of the vehicle, a first plurality of predicted vehicle states of the vehicle at a future point in time;
obtaining, from a vehicle planning system of the vehicle, a first candidate trajectory for the vehicle to follow, the first candidate trajectory being one of a plurality of candidate trajectories to be evaluated;
generating, by the generative model, a second plurality of predicted vehicle states based on the first candidate trajectory, the second plurality of predicted vehicle states representing predicted vehicle states at the future point in time assuming that the vehicle has initiated the first candidate trajectory generated by the vehicle planning subsystem, wherein the second plurality of predicted vehicle states are based on a set of characteristics for the vehicle after initiating the first candidate trajectory;
generating a measure of observer surprise for the observer of the vehicle including comparing the first plurality of predicted vehicle states to the second plurality of predicted vehicle states based on the first candidate trajectory, the measure of observer surprise representing how surprising the vehicle following the first candidate trajectory would be to the observer; and
based on the measure of observer surprise, causing the vehicle to select and follow a different candidate trajectory from the plurality of candidate trajectories.

12. The system of claim 11 , wherein the generative model is trained to generate predicted vehicle states from the perspective of a passenger of the vehicle.

13. The system of claim 11 , wherein the generative model is trained to generate predicted vehicle states from the perspective of another vehicle.

14. The system of claim 11 , wherein the generative model is trained to generate predicted vehicle states from the perspective of a pedestrian.

15. The system of claim 11 , wherein the generative model is trained to generate predicted vehicle states from the perspective of a bicyclist.

16. The system of claim 11 , wherein comparing the first plurality of predicted vehicle states to the second plurality of predicted vehicle states comprises determining a Kullback-Leibler divergence between a first probability distribution representing the first plurality of predicted vehicle states and a second probability distribution representing the second plurality of predicted vehicle states.

17. The system of claim 11 , wherein the first candidate trajectory includes a geometry component and a speed profile that define a future behavior of the vehicle, and wherein the measure of observer surprise corresponds to how surprising the future behavior is to the observer.

18. The system of claim 11 , wherein the observer is a road user, and the method further comprises:
receiving sensor data identifying characteristics of the observer; and
using the generative model to predict a future behavior of the road user, and wherein generating the second plurality of predicted vehicle states is further based on the predicted future behavior of the road user.

19. The system of claim 11 , wherein the operations further comprise:
using the vehicle planning system to generate the different candidate trajectory for the vehicle; and
determining a second measure of observer surprise for the different candidate trajectory, and wherein causing the vehicle to select the

Assignees

• Waymo Llc

Inventors

• Engstrom Johan
• Russell Jared

Classifications

• B60W60/0017 Primary

  of other traffic participants · CPC title

• B60W50/0098 Primary

  Details of control systems ensuring comfort, safety or stability not otherwise provided for · CPC title

• G05B13/048

  using a predictor · CPC title

• B60W2400/00

  Indexing codes relating to detected, measured or calculated conditions or factors · CPC title

• B60W2050/0014

  Adaptive controllers · CPC title